What Did Teachers’ Unions Do During the COVID-19 Pandemic? Evidence Based on Returns on Teacher Unionization

Abstract

This study investigates the impact of the COVID-19 pandemic on teachers’ labor market outcomes and estimates the returns on unionization. Using a difference-in-differences approach, I identify the effects of the pandemic on employment, earnings, and other labor market outcomes for unionized teachers relative to their non-unionized counterparts. The findings suggest that unionized teachers experienced greater job security and maintained their pre-pandemic wage premium. The role of unions varies significantly across teacher characteristics. Additionally, unionized teachers were more likely to work remotely and remain employed during the pandemic.

The COVID-19 pandemic in 2020 posed unprecedented challenges to the U.S. educational system. Nationwide school closures left many teachers adrift. As schools began reopening in the fall of 2020, teachers faced difficult working conditions. They had to adapt quickly to new instructional modes—including in-person, remote, and hybrid teaching—while also navigating serious health concerns, as many were required to return to classrooms at the height of the pandemic. These compounded struggles echoed and intensified the momentum of large-scale teacher strikes that began in 2018. Notably, the six major teacher strikes recorded in 2021–2022 represent one of the highest counts in the past three decades [1].

Most of these strikes were organized by teachers’ unions, which have historically played a critical role in securing meaningful gains for educators. Typical union objectives include advocating for improved employment conditions (such as higher salaries and safer working environments), increased resources for professional development, greater classroom autonomy, and overall improvements to teachers’ well-being [2,3,4,5]. Moreover, teachers’ unions serve as a collective voice for educators, enabling them to express workplace concerns and influence policy [6,7]. Given this context, a key question emerges: What role did teachers’ unions play during the COVID-19 pandemic?

This research aims to examine the impact of the pandemic on the labor market outcomes of unionized versus non-unionized public school teachers. Since teachers’ unions pursue multiple goals related to members’ well-being, it is important to understand which aspects of employment they were able to influence during the crisis. Specifically, the study addresses the following research questions: (1) Did teachers’ unions help protect members’ job security during the COVID-19 pandemic? (2) How did the pandemic affect the union wage premium among public school teachers? (3) What was the impact of COVID-19 on other labor market conditions for unionized teachers? (4) How did these effects vary by teacher characteristics?

Using data from the Current Population Survey (CPS) from 2015 to 2021, I employ a difference-in-differences approach to estimate the pandemic’s effects on public school teachers’ employment outcomes. The findings indicate that unionized teachers experienced greater job security during the pandemic relative to their non-unionized peers, while maintaining the union wage premium observed before the pandemic. Results also reveal substantial heterogeneity in union effects across different teacher subgroups. Additionally, unionized teachers were more likely to work remotely and to remain employed during the pandemic.

This study contributes to the literature on the role of teachers’ unions in shaping compensation and employment outcomes by leveraging nationally representative data and incorporating pandemic-specific labor market indicators. It also offers a more nuanced understanding of how union effects vary across teacher subgroups, shedding light on the diverse ways unions supported their members during a major public crisis.

1. Literature

1.1. Teachers’ Unions and Employment During the COVID-19 Pandemic

During the COVID-19 pandemic, various factors contributed to less severe job loss among union workers. First, during an economic downturn, unions are often able to arrange work sharing, encouraging employers to temporarily reduce the work hours of their employees instead of laying them off, in order to save jobs and avert layoffs [8,9,10,11].

Second, unions also serve as workers’ “voice” in the workplace. Thus, compared to non-union workers, union workers are less worried about speaking out regarding safety concerns in their workplace because unions seek better protection and stronger safety measures on their behalf [7,12,13,14].

Third, union employers are more likely to offer health insurance and generous paid sick leave than non-union employers [15,16,17,18,19,20]. With better access to health care and medical treatment than their non-union counterparts, union workers are more likely to remain in their jobs when facing the same health risks.

Indeed, Han [21] finds that union workers, compared to non-union workers, had greater job security during the COVID-19 pandemic, and the union workers’ employment advantage was greater for marginalized workers (female and minority workers).

Nevertheless, whether the teacher labor market also reveals union employment benefits needs more in-depth investigation. As schools reopened after the nationwide closure during the pandemic, teachers came back to complicated instructional platforms and approaches. According to new survey data from the nationally representative RAND American Teacher Panel, during the pandemic, more teachers became unsure that they would remain in the classroom [22]. In March 2020, 74% of teachers reported that they expected to work as a full-time teacher until retirement, whereas in March 2021, that number decreased to 69%. A new survey from the National Education Association finds that more than half of American teachers say they are planning to leave teaching or retire early because of the COVID-19 pandemic [23]. Some teachers who experience long-term symptoms of COVID-19 also plan to retire early [24].

If teachers in union-friendly schools and districts were less likely to quit or to be laid off during the pandemic, partly because unionized districts received more funding, the teachers’ labor market would also reveal the employment advantages of unions. On the other hand, if unionized schools and districts tended to experience prolonged school closure or suffered more financial distress, teachers’ unions may have led to employment disadvantages. Thus, the overall impact of the pandemic on job security and the employment of unionized teachers relative to non-unionized teachers presents empirical questions.

1.2. Union Wage Premium Among Teachers During the COVID-19 Pandemic

The literature often finds countercyclical behavior of the union–non-union wage differential during a recession. Keane, Moffitt, and Runkle [25] point out that the selectivity can generate countercyclical behavior of the union wage premium. Rees [26] shows that the union–non-union wage gap widened during the 1974–1975 recession, as non-union wages either rose by a smaller amount or did not rise at all. The wage gap between unionized and non-unionized workers can increase during an economic downturn because high-wage workers tend to bear greater risks of layoffs in the non-union sector compared to the union sector. Also, high-wage workers with greater wealth can afford a lengthier job search period and remain unemployed longer than low-wage workers in the non-union sector [27,28]. Other studies cite the wage rigidity and seniority rules of unions during recessions as the reasons for the countercyclical nature of the union wage premium [29,30].

A theory of compensating wage differentials (CWDs) predicts that there exists a positive wage premium for health risk on the job, and numerous studies find higher wages as the risks on the jobs are greater [31,32,33]. Because the COVID-19 pandemic introduced a huge crisis to worker health, teachers’ unions may have been able to negotiate for higher pay when the school reopened, potentially increasing the union wage premium.

However, according to union rent-sharing models, unions attempt to lower wages to try to save jobs when employers are in trouble [34]. This model shows that unions whose members earn above-market wages should be willing to make concessions such that their companies can stay in business. Several studies show that, during the recession in 2008–2009, unions made sizable wage and benefit concessions to save jobs [35,36,37].

Public school teachers have relatively high job security compared to other occupations. During a severe recession, job security can be an additional benefit of teacher unionization. This appealing attribute of unions may be associated with reduced pay because unions may be willing to accept wage cuts to keep jobs during a pandemic, which is a negative CWD [38]. Conversely, as more teachers express their desire to leave teaching early, unions may be able to retain their wage premium during a pandemic.

Han [21] finds that the pre-pandemic union wage premium in the public sector remained largely unchanged during the pandemic-induced recession. Over half of union workers in the U.S. are in the public sector, and public school teachers make up the single largest group of public sector workers. Thus, it is important to examine what happened to the union wage premium for teachers during the pandemic-induced recession.

1.3. Heterogeneity of the Union Impact During the COVID-19 Pandemic

The pandemic effects on the relative employment and the pay of unionized teachers over non-unionized teachers may depend on teacher characteristics. There are several reasons why one may expect to observe different union influences on women and minorities. Studies on gender equality emphasize the importance of institutional support for unions and collective bargaining to represent inclusive gender interests. Refs. [39,40,41,42] show that some unions adopt gender-specific strategies to recruit females by emphasizing issues of particular interest to women, such as work–family balance and workplace discrimination. For instance, because teaching is a female-dominated profession, teachers’ unions are aware of the benefits of paid leaves. Budd and Brey [15] suggest that minority workers could benefit more from union representation, which helps them to obtain paid family and medical leave.

During the recession, family relations may also be an important factor for union influence in the labor market. According to Kochhar [43] from the Pew Research Center, the shares of working mothers and fathers with children at home were significantly reduced, while the shares of parents not in the labor force (neither working nor actively looking for jobs) were increased due to the COVID-19 pandemic. Moreover, the employment of working parents with children younger than 5 years old was more negatively affected by the pandemic compared to parents whose youngest child was 14 to 17 years old. Because union workers are more likely to receive health insurance and generous paid leaves, parents with children were better able to maintain their employment during the pandemic.

Building on this literature, I conduct separate analyses for various subgroups of teachers to address individual heterogeneity in assessing the effect of the pandemic on unionized teachers relative to non-unionized teachers.

1.4. Teachers’ Unions and Other Labor Market Conditions During the COVID-19 Pandemic

There were other labor market conditions that unionized workers were more likely to face during the pandemic. Golden [44] finds that unions are associated with less access to flexible work schedules. Wallender [45] analyzes more than 4300 documents in Bloomberg Law’s online library of collective bargaining agreements and finds that less than 4% of union contracts over the past five years mention “telecommuting”, “telework”, or “work from home”. Similarly, “flexible work scheduling” has not been a popular item in union contracts. This may be due to the intrinsic features of many union jobs (in manufacturing, construction, healthcare, and education, for instance), where the essential part of the job requires workers’ physical presence. Thus, union employers prefer to keep their employees working on site. Therefore, during the COVID-19 pandemic, union employers may have been less likely to move towards remote working compared to non-unionized employers.

However, the teachers’ labor market faces a rather different environment in terms of remote working. During the school closure in 2020, most teachers were required to teach fully remotely, and many teachers used hybrid models as schools reopened. Teachers’ unions often requested remote learning citing the public health risks of in-person instruction. Sawchuk [46] reports that when teachers’ unions are renegotiating their terms with their districts, a discussion on remote learning is on the table. Several studies find that districts and states with strong union power were less likely to reopen schools, were less likely to have in-person instruction, and spent fewer weeks in in-person learning [47,48,49,50].

If teachers’ unions are more likely to bargain for in-person learning, more unionized teachers (either union members or non-members who are covered by union contracts) would work remotely than non-unionized teachers.

1.5. International Studies on Public Education and Teachers’ Unions During COVID-19

Internationally, trade unions played a pivotal role in responding to the COVID-19 crisis, consistently emphasizing the active involvement of workers in social dialogue. These efforts were central to advancing labor and social agendas aimed at fostering resilience and empowerment in the face of global disruptions [51]. Within the public education sector, the emerging literature highlights significant increases in teacher stress and burnout during the pandemic, which have contributed to worsening rates of teacher attrition and exacerbated longstanding teacher shortages [52,53,54,55]. According to the Forward to School report by Education International [56], teachers’ unions across various countries have actively advocated for enhanced support systems for educators, including resources targeting mental health and overall well-being. Moreover, European educators’ unions have specifically underscored the disproportionate challenges faced by teachers and students with disabilities, calling for expanded access to health services and social supports in light of persistent informational and structural barriers [57].

2. Data

This study draws on data from the Current Population Survey (CPS). The CPS is a monthly household survey administered by the U.S. Bureau of Labor Statistics (BLS). The CPS’ basic monthly survey is the primary source of data on labor force characteristics as well as demographic information on the U.S. population. The BLS surveys more than 65,000 households, which results in 90,000–120,000 individual observations per month. Every household that enters the CPS is interviewed each month for 4 months, with a pause for 8 months, and then interviewed again for 4 more months. Thus, we observe the same individuals in the same household for two consecutive years provided that they do not move.

Household members that are interviewed for the fourth or eighth month are asked additional questions, such as regarding union membership, weekly earnings, hourly wage, and if they are 16 years and older and currently employed as a wage or salaried worker (this is also known as the “earner study” of the CPS). This implies that only employed workers are eligible to answer the union membership question. These outgoing interviews are the only ones included in the extracts of the Outgoing Rotation Group (ORG). The ORG households, which typically comprise one-fourth of every basic monthly sample for a calendar year, are extracted and merged by the Census Bureau and are released as the Merged Outgoing Rotation Group file (MORG). I also utilized these MORG files, retrieved from the IPUMS-CPS [58].

I then dropped workers who did not report earnings but instead had them imputed from the estimation samples, because including these imputed earners could lead to under-estimation of union wages and over-estimation of non-union wages, resulting in underestimation of union–non-union wage differentials [59].

In May 2020, the BLS included a series of questions related to the COVID-19 pandemic at the end of the basic monthly questionnaire. I selected the following two questions to measure the effects of the COVID-19 pandemic on other labor market conditions:

Q1. At any time in the last 4 weeks, did you telework or work at home for pay because of the coronavirus pandemic? (Yes or No)

Q2. At any time in the last 4 weeks, were you unable to work because your employer closed or lost business due to the coronavirus pandemic? (Yes or No)

After restricting the sample to public school teachers (elementary, middle, high school, special education teachers, and other teachers and instructors working in the public sector) with at least a bachelors’ degree, I pooled monthly data from 2015 through 2021. Appendix A provides the summary statistics of the selected variables from the pooled CPS for 2015–2021.

I defined a binary indicator for labor unions based on whether a teacher is a member of labor unions or not a union member but covered by a union contract. In

Table 1. Mean characteristics of public school teachers by union status.

	Union Status
	Union (SD)	Non-Union (SD)	Difference (SE)
Male	0.216 (0.412)	0.215 (0.411)	0.001 (0.014)
Age	44.93 (13.14)	43.51 (15.86)	1.412 *** (0.439)
Full time	0.629 (0.483)	0.426 (0.494)	0.203 *** (0.016)
Master’s degree and above	0.472 (0.499)	0.256 (0.436)	0.216 *** (0.015)
Citizen	0.982 (0.133)	0.970 (0.171)	0.012 ** (0.006)
Veteran	0.017 (0.128)	0.043 (0.208)	−0.026 *** (0.005)
Married	0.632 (0.482)	0.569 (0.495)	0.063 *** (0.017)
Children under the age 5	0.143 (0.427)	0.110 (0.383)	0.033 *** (0.013)
Race and ethnicity
White	0.808 (0.393)	0.770 (0.420)	0.038 *** (0.014)
Black	0.135 (0.341)	0.148 (0.335)	−0.013 (0.012)
Asian	0.026 (0.161)	0.041 (0.199)	−0.015 ** (0.006)
Hispanic	0.153 (0.359)	0.129 (0.334)	0.024 ** (0.012)
Metropolitan area	0.808 (0.393)	0.685 (0.464)	0.123 *** (0.013)
School level
Elementary and middle school	0.632 (0.482)	0.584 (0.492)	0.048 *** (0.016)
Secondary school teachers	0.175 (0.380)	0.125 (0.331)	0.050 *** (0.012)
Special education teachers	0.119 (0.323)	0.066 (0.249)	0.052 *** (0.009)
Other teachers and instructors	0.073 (0.262)	0.223 (0.416)	−0.150 *** (0.011)
Observations	1895	2921

Notes: *** p < 0.01, ** p < 0.05. Sample includes public school teachers who are college graduates. Observations refer to CPS sample size. Union refers to a member of union or non-member but covered by union contracts. Numbers are weighted by workers’ earnings weights. Source: CPS Basic combined with ORG, 2015–2021.

, I compare the mean characteristics by teacher’s union status. On average, compared to non-unionized teachers, unionized teachers are more likely to be older, working full time, more educated, married, have children under the age of 5, and residing in metropolitan areas. In terms of race/ethnicity, white and Hispanic teachers are more likely to join unions. On average, elementary and middle school teachers have the highest union membership rates among all school levels.

The information concerning union status is not queried unless a worker is employed, so I could only investigate employment responses to the pandemic by aggregating the employment numbers to the state-month-year level. Based on the monthly panel data by state, Figure 1 describes how the monthly employment differs by teachers’ union status between 2015 and 2021. For both unionized and non-unionized teachers, there exists a large fluctuation, reflecting the cyclical nature of employment (i.e., summer and winter breaks of an academic year). After March 2020, the number of teachers who are employed, both unionized and non-unionized, substantially drops to the lowest point, following school closure. Starting in August–September 2020, the employment of teachers picks up as schools reopen for the fall semester. As of December 2021, teachers’ employment had not yet returned to the pre-pandemic levels. Based on this employment trend, I define the pandemic trough over the first six months of the COVID-19 outbreak, between March 2020 and August 2020, and the recovery period after September 2020. Figure 1 does not exhibit a noticeable difference in the employment trend between unionized and non-unionized teachers, possibly because it illustrates the unconditional mean of employment of teachers from both groups.

Figure 2 depicts the time trend for teachers’ real weekly earnings (truncated at the bottom and top 1% and adjusted to 2015 USD) by union status. There exists a consistent union wage premium in pre-pandemic periods. In mid-2019, the union–non-union wage gap seems to shrink, as the wages of non-unionized teachers sharply increase, partly as a result of successful teacher strikes countrywide. Nonetheless, teachers’ union wage premium remains significant right up to the onset of the COVD-19 pandemic. Since the pandemic, the real earnings of unionized teachers show an increasing pattern, whereas the real earnings of non-unionized teachers are relatively stagnant.

Figure 3 presents the mean comparison of other labor market conditions between unionized and non-unionized teachers, based on the new pandemic-related questionnaires. Overall, unionized teachers, compared to non-unionized teachers, were more likely to work remotely. Non-unionized teachers are more likely to report that they were unable to work due to employer closure. The differences between unionized and non-unionized teachers for both items are statistically significant at the 1% significance level.

3. Empirical Strategy

The COVID-19 pandemic created an exogenous shock to the labor market on a global scale. To identify the impact of the COVID-19 pandemic on the labor market outcomes of unionized teachers relative to non-unionized teachers, I estimate the following the difference-in-difference (DID) model, using unionized teachers as the treatment group and non-unionized teachers as the comparison group, based on state-monthly panel data:

$$\mathrm{Log}({\mathrm{Emp})}_{\mathrm{sm}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{Union}}_{\mathrm{s}}+{\mathsf{\alpha }}_2{\mathrm{Covid}}_{\mathrm{m}}+{\mathsf{\alpha }}_3\left({\mathrm{Union}}_{\mathrm{s}}\times {\mathrm{Covid}}_{\mathrm{m}}\right)+{\mathrm{X}}_{\mathrm{sm}}+{\mathsf{\lambda }}_{\mathrm{s}}+{\mathsf{\tau }}_{\mathrm{m}}+{\mathsf{\epsilon }}_{\mathrm{sm}},$$

(1)

where s and m indicate state and month, respectively. Emp represents the number of teachers employed. Union equals 1 for the treatment group (unionized teachers) and 0 for the control group (non-unionized teachers), and Covid equals 1 if the time period follows the COVID-19 pandemic (March 2020 or later) and 0 otherwise. The coefficient of the interaction term between Union and Covid, α₃, is the DID estimator, which estimates the difference in employment between unionized and non-unionized teachers before and after the COVID-19 pandemic. All models are weighted by the teachers’ earnings weight, which is a proper weight whenever the MORG variables are used, and standard errors are clustered at the state level.

To improve the baseline DID estimates, I include control variables ($X_{sm}$) to reduce potential omitted variable bias. The covariates I include are the monthly averages of the following variables: teacher age, usual work hours, number of children aged 5 and under, proportion of workers who are male, full-time workers, veterans, citizens, and categorical variables for race, ethnicity, marital status, metropolitan area, and school level.

I also add the state fixed-effects (${\lambda }_s$) to control for time-invariant confounding factors that are specific to each state, such as demographics, industrial structure, economic background, and cultural/political environment towards public education. I also include the monthly fixed-effects (${\tau }_m$) to control for national time trends for macroeconomic conditions, federal holidays, and the common pandemic timeline that hit all states, including nationwide school closures and economic relief by the federal government.

To estimate the effect of COVID-19 on the earnings of unionized teachers relative to the earnings of non-unionized teachers, I use the DID estimation of the following equation:

$${\mathrm{Log}\left(\mathrm{Wage}\right)}_{\mathrm{im}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{Union}}_{\mathrm{i}}+{\mathsf{\alpha }}_2{\mathrm{Covid}}_{\mathrm{m}}+{\mathsf{\alpha }}_3\left({\mathrm{Union}}_{\mathrm{i}}\times {\mathrm{Covid}}_{\mathrm{m}}\right)+{\mathrm{X}}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{s}}+{\mathsf{\tau }}_{\mathrm{m}}+{\mathsf{\epsilon }}_{\mathrm{im}},$$

(2)

where i and m indicate individuals and months, respectively. Wage represents the real weekly earnings adjusted to 2015 USD and truncated at the top and bottom 1%. The coefficient α₃ is the DID estimator, which estimates the difference in real weekly earnings between unionized and non-unionized teachers before and after the pandemic.

For the augmented DID estimation, I control for teachers’ gender, age, usual work hours, full-time status, veteran status, citizenship status, number of children aged 5 and under, with categorical variables for education, race, ethnicity, marital status, metropolitan area, and school level. All models are weighted by teachers’ earnings weight, and standard errors are clustered at the state level. I also add state, month, and year dummies.

The DID results may not be driven by the pandemic impact but by systematic differences in treatment (unionized teachers) and control groups (non-unionized teachers). Thus, the key assumption for the DID estimator is that the treatment and control groups would have parallel trends in employment and real wages before the pandemic. The parallel trends assumption states that, in the absence of treatment, the treated and control groups would have exhibited similar trends in their outcomes over time. This ensures that any post-treatment differences between the groups can be attributed to the treatment effect rather than to pre-existing differences in their trajectories. Although we cannot observe the counterfactual—what would have happened to the outcome variables for the treatment group if the pandemic had not occurred—we can examine the trends in outcome variables for both groups before the pandemic and examine whether the two groups are indeed comparable. To examine whether this assumption is likely to hold, I use an event-study approach of the following equation:

$$Y_{sm}={\beta }_0+{\sum }_{k=-14}^{-2}{\beta }_k{Event}_{sk}+{\sum }_{k=0}^{21}{\beta }_k{Event}_{sk}+X_{sm}+{\lambda }_s+{\tau }_m+{\epsilon }_{im},$$

(3)

where Event is the number of periods since the COVID-19 pandemic outbreak (March 2020), which is normalized to −1 for control units. Y represents the employment gap or wage gap between unionized and non-unionized teachers. X is the vector of control variables, which includes the same set of variables used in Equation (1) for the employment gap and in Equation (2) for the wage gap between unionized and non-unionized teachers. ${\lambda }_s$ represents state dummies and ${\tau }_m$ represents month dummies. Each of the event study coefficients, β_k, is a simple DID estimator, using the period just before the onset of the pandemic as the “before” period and the period of each event study coefficient as the “after” period. Thus, coefficient β_k measures the difference in the outcome variables between treatment and control groups at a specified period, relative to the difference between the two groups one month before the pandemic. If the pre-pandemic trends of the employment and wage gaps between unionized and non-unionized teachers are constant, β_k would be close to zero for any period before the pandemic. If the pandemic improved the employment status and real wages of unionized teachers relative to non-unionized teachers, β_k would be positive for any period after the pandemic.

Figure 4 presents the testing of the parallel-trend assumption of the DID estimation for monthly employment (in Figure 4a) and the monthly average of real wages (in Figure 4b) using the event-study approach. The estimated coefficients (β_k in Equation (3)) and their 95% confidence intervals are plotted, and the vertical lines refer to the pandemic outbreak. The β_k in both figures before the pandemic outbreak oscillates around zero, indicating that the pre-treatment trends for both employment and real weekly earnings are constant for unionized and non-unionized teachers. This provides evidence that the parallel-trend assumption is likely to be satisfied and that my estimation is based on two comparable groups.

To capture possible shifts in the union–non-union differentials over time post-pandemic, I analyze the data using two different “after” periods, based on the timeline of the pandemic: the pandemic trough (March 2020–August 2020) and the recovery period (September 2020–December 2021). In addition, to examine how the impact of the COVID-19 pandemic differs by teacher characteristics, I analyze the data for various teacher subgroups.

To examine the role of teachers’ unions in other Covid-related labor market conditions, I run the following OLS regression:

$$Y_{imst}={\beta }_0+{\beta }_1{Union}_{imst}+{\beta }_2{X}_{imst}+{\delta }_s+{\lambda }_t+{\epsilon }_{imst,},$$

(4)

where i, m, s, and t indicate individual, month, state, and year, respectively. $Y$ represents the new COVID-19-related items presented in Figure 2. Union measures union status. X is the vector of control variables. ${\delta }_s$ represents state dummies, ${\lambda }_t$ is the year dummy, and $\epsilon$ is the error term reflecting variation not accounted for in the regression model.

4. Results

Table 2. Impact of COVID-19 on employment of teachers by union status.

Log (Monthly Employment)	Before	After	Difference = After − Before
Panel A: Pandemic Trough	January 2015–February 2020	March 2020–August 2020
Union	3.163 (0.029)	2.908 (0.100)	−0.254 *** (0.102)
Non-union	3.646 (0.027)	3.036 (0.139)	−0.610 *** (0.141)
Difference = Union − Non-union	−0.483 *** (0.041)	−0.127 (0.159)	0.356 ** (0.156)
Panel B: Recovery Period	January 2015–February 2020	September 2020–December 2021
Union	3.163 (0.029)	2.821 (0.049)	−0.341 *** (0.057)
Non-union	3.646 (0.027)	3.080 (0.072)	−0.566 *** (0.077)
Difference = Union − Non-union	−0.483 *** (0.041)	−0.259 ** (0.086)	0.225 ** (0.095)

Notes: Errors are clustered within states (presented in parenthesis). *** p < 0.01, ** p < 0.05. Sample includes public school teachers who are college graduates. Union refers to a member of union or non-member but covered by union contracts. The models are weighted by earnings weights. Source: Monthly panel data by union status and state, based on CPS basic combined with ORG, 2015–2021.

presents the results from the baseline DID estimation for monthly employment by summarizing the impact of the COVID-19 pandemic on the log of employment of unionized teachers and non-unionized teachers. Panel A focuses on the pandemic trough, and panel B on the recovery period as the “after” period. Both panels show that the pandemic had a negative impact on the employment of both unionized and non-unionized teachers, but the magnitude of its impact was much greater for non-unionized teachers. Panel A shows that because unionized teachers experienced significantly less job loss than the non-unionized teachers, the employment gap between the unionized and non-unionized teachers substantially narrowed during the pandemic trough. This resulted in a relative employment gain of 35.6% of the unionized teachers, which is the estimated DID estimator of α₃ in model (1). In Panel B, the DID estimate of the employment advantage of teacher unionization during the recovery period is 22.5%, which is smaller than the estimate during the pandemic trough, indicating that the unions’ employment benefits become smaller as the economy tried to bounce back.

Figure 4a describes a consistent story. The estimated coefficients are around zero before the pandemic, but they turn positive after the pandemic, showing that unionized teachers, compared to non-unionized teachers, enjoyed greater job security during the pandemic. The magnitude of the unions’ employment benefit is greater during the pandemic trough than the recovery period. I also re-estimate the event study model with the demographic information from the augmented DID estimation. The inclusion of these control variables does not change the pre-pandemic estimates.

The baseline DID estimates presented in Table 2 do not account for any differences in teacher characteristics that may be associated with union status. In an attempt to solve this problem, I control for the monthly averages of the variables presented in Table 1.

Table 3. Impact of COVID-19 on relative employment of union teachers by subgroup. Augmented DID.

Log (Monthly Employment)	Panel A: Pandemic Trough (After: March 2020–August 2020)	Panel B: Recovery Period (After: September 2020–December 2021)
Section A: All	0.410 ** (0.201)	0.185 * (0.101)
Section B: Gender
Male	0.958 ** (0.470)	0.354 ** (0.180)
Female	0.373 ** (0.185)	0.169 * (0.102)
Section C: Race and Ethnicity
White	0.404 * (0.208)	0.167 * (0.100)
Black	0.184 (0.290)	0.350 (0.272)
Asian	0.713 * (0.372)	0.041 (0.505)
Hispanic	0.702 ** (0.279)	0.106 (0.247)
Section D: School Level
Elementary and Middle School	0.730 *** (0.187)	0.255 ** (0.126)
Secondary School Teachers	0.123 (0.369)	0.488 ** (0.233)
Special Education Teachers	−0.260 (0.454)	−0.329 (0.228)
Other Teachers and Instructors	0.273 (0.395)	0.030 (0.227)
Section E: Experience
Early-career Teachers (exp ≤ 5)	0.218 (0.416)	0.028 (0.273)
Mid-career Teachers (6 ≤ exp ≤ 20)	0.190 (0.290)	0.117 (0.197)
Senior Teachers (exp > 20)	0.526 *** (0.200)	0.279 ** (0.131)
Section F: Children
Children under the age 5	−0.499 (0.367)	0.055 (0.297)
No children under the age 5	0.415 ** (0.209)	0.187 * (0.111)

Notes: Errors are clustered within states (presented in parenthesis). *** p < 0.01, ** p < 0.05, * p < 0.1. Public school teachers who are college graduates are included in the sample. Union refers to a member of union or non-member but covered by union contracts. The models are weighted by earnings weights. Covariates include the monthly averages of the following variables: workers’ age, number of children aged 5 and under, proportion of workers who are male, full-time workers, veterans; categorical dummy variables include education level, race, ethnicity, marital status, citizenship status, school level, metropolitan area, state, year, and month. Covariates are omitted, as appropriate, for estimates of different subsamples. Source: CPS Basic combined with ORG, 2015–2021.

summarizes the estimated coefficient α₃ in model (1)—the augmented DID results—which shows the change in employment gap (in log) between unionized and non-unionized teachers due to the pandemic, after controlling for a full set of covariates, by teacher subgroup.

Panel A of Table 3 reports the augmented DID results for the pandemic trough A, with Panel B for the recovery period. Section A in Panel A shows that, overall, in the first 6 months of the pandemic, teachers’ unions saved 41 percent more jobs per month, which is greater than the baseline estimate presented in Table 2. Section A in Panel B shows that unionized teachers lost fewer jobs than non-unionized teachers, by 18.5 percent, during the recovery period. These results suggest that teacher characteristics play essential roles in the different employment outcomes of unionized and non-unionized teachers, and controlling for these factors is critical in assessing the impact of the pandemic on the employment gap between the union and non-union sectors.

Sections B through E in Table 3, which present the DID estimates by various subgroup of teachers, show that there exists a large heterogeneity in the unions’ role by teacher characteristic. In Panel A, during the pandemic trough, male teachers were bigger beneficiaries of the unions compared to female teachers. Black teachers did not experience unions’ employment advantage, whereas unionized teachers from other racial groups enjoyed better job security. Union teachers in elementary and middle schools received a large employment gain. Senior unionized teachers were better able to keep their jobs compared to less experienced teachers. Teachers without young children were able to receive unions’ employment benefit. Panel B of Table 3 shows similar patterns of union role but with less magnitude during the recovery period. Union teachers who were male, white, teaching grades K–12, senior, and with no small children had a greater job security.

Whether teachers were able to maintain the union wage premium during the pandemic can only be answered through empirical analysis.

Table 4. Impact of COVID-19 on real earnings of teachers by union status.

Log (Real Weekly Earnings)	Before	After	Difference = After − Before
Panel A: Pandemic Trough	January 2015–February 2020	March 2020–August 2020	Diff-in-Diff
Union	6.664 (0.017)	6.660 (0.068)	−0.004 (0.070)
Non-union	6.035 (0.018)	6.162 (0.074)	0.127 * (0.076)
Difference = Union − Non-union	0.629 *** (0.040)	0.498 *** (0.121)	−0.131 (0.146)
Panel B: Recovery Period	January 2015–February 2020	September 2020–December 2021	Diff-in-Diff
Union	6.664 (0.017)	6.894 (0.035)	0.230 *** (0.039)
Non-union	6.035 (0.018)	6.153 (0.044)	0.118 ** (0.047)
Difference = Union − Non-union	0.629 *** (0.040)	0.741 *** (0.074)	0.112 (0.079)

Notes: Errors are clustered within states (presented in parenthesis). *** p < 0.01, ** p < 0.05, * p < 0.1. Public school teachers who are college graduates are included in the sample. Weekly earnings are truncated at the top and bottom 1% and adjusted to 2015 USD. The models are weighted by earnings weights. Source: CPS Basic combined with ORG, 2015–2021.

presents the results from the baseline DID estimation on real weekly earnings. In Panel A, unionized teachers’ wages did not change, whereas non-unionized teachers received wage increases during the pandemic trough. This led to a 13.1% decrease in the union wage premium. Panel B shows that both unionized and non-unionized teachers received significant wage gains as the economy recovered, with unionized teachers showing a slightly bigger raise than non-unionized teachers (23.0% vs. 11.8%). Thus, union wage premium increased by 11.2%. However, in both panels, the DID estimators are not statistically significant, suggesting that the pandemic had a null impact on the union wage premium during the course of the pandemic.

Figure 4b depicts the estimated coefficients, which fluctuate around zero both before and after the pandemic. This also shows that the pandemic did not influence the earnings gap between unionized and non-unionized teachers, and unionized teachers were able to maintain their wage premium during the pandemic.

Table 5. Impact of COVID-19 on relative labor earnings of union teachers by subgroup. Augmented DID.

Log (Real Weekly Earnings)	Panel A: Pandemic Trough (After: March 2020–August 2020)	Panel B: Recovery Period (After: September 2020–December 2021)
Section A: All	−0.090 (0.105)	0.097 * (0.058)
Section B: Gender
Male	−0.232 (0.226)	0.230 * (0.124)
Female	−0.080 (0.143)	-0.033 (0.057)
Section C: Race and Ethnicity
White	−0.145 (0.128)	0.089 * (0.054)
Black	−0.050 (0.215)	−0.138 (0.162)
Asian	0.711 * (0.354)	0.039 (0.213)
Hispanic	0.102 (0.112)	−0.074 (0.117)
Section D: School Level
Elementary and Middle School	−0.063 (0.172)	0.003 (0.077)
Secondary School Teachers	−0.449 * (0.269)	−0.089 (0.117)
Special Education Teachers	0.558 ** (0.258)	1.885 * (0.126)
Other Teachers and Instructors	−0.473 * (0.229)	0.239 (0.156)
Section E: Experience
Early-career Teachers (exp ≤ 5)	0.149 (0.262)	0.265 * (0.143)
Mid-career Teachers (6 ≤ exp ≤ 20)	−0.201 (0.213)	0.112 (0.099)
Senior Teachers (exp > 20)	−0.054 (0.111)	−0.014 (0.070)
Section F: Children
Children under the age 5	−0.208 (0.280)	0.063 (0.148)
No children under the age 5	−0.071 (0.180)	0.047 (0.059)

Notes: Errors are clustered within states (presented in parenthesis). ** p < 0.05, * p < 0.1. Public school teachers who are college graduates are included in the sample. Weekly earnings are adjusted to 2015 USD and truncated at the top and bottom 1%. The models are weighted by earnings weights. Covariates include gender, age, education, race/ethnicity, work hours, veteran status, number of kids aged 5 and under, marital status, citizenship status, school level, metropolitan area, sector, industry, state, year, and month dummies. Covariates are omitted, as appropriate, for estimates of different subsamples. Source: CPS Basic combined with ORG, 2015–2021.

summarizes the augmented DID estimates after controlling for various teacher characteristics by the pandemic timeline. In Panel A, overall, the union wage premium was largely intact during the pandemic trough. On average, the union wage premium fell, but the decrease was not statistically significant. Sections B through E show mixed effects of the pandemic on the union wage premium in various subgroups of teachers. Asian teachers and special education teachers experienced an increase in the union wage premium, whereas secondary school teachers saw a decrease in the union wage premium. For other subgroups of teachers, the changes in wages of unionized teachers are not statistically significantly different from the changes in wages of non-unionized workers after the onset of the pandemic.

During the recovery period in Section A of Panel B, however, unionized teachers earned a 9.7% higher wage premium compared to the pre-pandemic period. Sections B through E show that unionized teachers who were male, white, teaching special education, and in their early career path saw an increased union wage premium as the labor market bounced back.

Teachers’ unions may also influence other labor market conditions besides employment and wages.

Table 6. The relationship between teachers’ unions and remote work during COVID-19.

	Panel A: Remote Working			Panel B: Unable to Find Work
VARIABLES	(1)	(2)	(3)	(1)	(2)	(3)
Labor Unions	0.174 ***	0.145 ***	0.139 **	−0.055 **	−0.057 **	−0.071 **
	(0.0628)	(0.0491)	(0.0531)	(0.0226)	(0.0230)	(0.0328)
Controls	X	X	X	X	X	X
Period dummies		X	X		X	X
State dummies			X			X
Observations	604	604	604	604	604	604
Adjusted R2	0.107	0.322	0.411	0.133	0.155	0.245

Notes: Errors are clustered within states (presented in parenthesis). *** p < 0.01, ** p < 0.05. Public school teachers who are college graduates are included in the sample. The models are weighted by final basic weights. Covariates include gender, age, education, race/ethnicity, full-time status, work hours, number of kids aged 5 and under, veteran status, marital status, citizenship, sector, industry, metropolitan areas, state, year, and month dummies. Source: CPS Basic combined with ORG, 2020 and 2021. In May 2020, the BLS added to the basic monthly questionnaire a series of questions related to COVID-19 and its impact on households.

presents the estimated relationship between unions and the COVID-19-related items shown in Figure 3. Based on my favored model specification in (3) in Panel A, unionized teachers were 14 percentage-points more likely to work remotely compared to non-unionized teachers. This result is in opposition to Han [21], who finds that union workers, compared to workers with no association with labor unions, were about 10 percentage-points less likely to work from home during the COVID-19 pandemic. This reveals that there exists heterogeneity in unions’ role depending on specific labor market characteristics.

Model (3) in Panel B shows that unionized teachers are 7.1 percentage-points less likely to report that they were unable to work due to employer closure. This suggests that, during the pandemic, unionized districts were less likely to face school closures, which may have led to a better employment status for unionized teachers.

The results in this study show that teachers’ unions influenced teachers’ labor market outcomes in various ways during the pandemic-induced recession. Compared to non-unionized teachers, unionized teachers enjoyed more flexibility in their work schedule via telework and were more likely to find work. This implies that teachers’ unions provide their members with employment stability and economic buffers during economic hardship.

5. Discussion and Conclusions

This research examines the impact of the COVID-19 pandemic on the returns of unionization of public school teachers. Based on nationally representative data, I employ the difference-in-difference estimation, using unionized teachers as the treatment group and non-unionized teachers as the comparison group, to estimate the effect of the pandemic on the employment and earnings gap between unionized and non-unionized teachers. I also compare other labor market conditions for unionized teachers relative to non-unionized teachers.

I find that, compared to non-unionized teachers, unionized teachers experienced a higher level of job security during the pandemic. Overall, the pre-pandemic level of the union wage premium remained largely unchanged during the pandemic trough, but the premium significantly increased during the recovery period. The unions’ advantage in employment and real wages greatly varies by teacher characteristic. So, further studies are warranted to understand the sources of the heterogeneous union influence on subgroups of teachers.

I also find that, compared to non-unionized teachers, unionized teachers were more likely to work remotely and to find work when facing employer closure during the pandemic. Remote working offers more flexibility to working parents, especially for mothers and elderly teachers. If teachers can work from home, more schools can reopen or remain open, potentially contributing to lower teacher turnover.

The findings of this study show that the overall labor market performance was far better for unionized teachers than for non-unionized teachers during the pandemic-induced recession. Union teachers enjoyed better job security while maintaining their pre-pandemic wage premium levels. Teachers’ unions played important roles in promoting teachers’ well-being by providing them with employment stability and flexibility as well as financial advantages in the absence of government intervention during the economic downturn.

There has been a strong anti-union sentiment and growing resistance to labor unions in the U.S. public sector. States such as Indiana, Kentucky, West Virginia, and Wisconsin are among the most recent to adopt right-to-work laws—bringing the total to 26. These laws allow employees to opt out of union membership while enjoying the same benefits of collective bargaining agreement in the same bargaining unit, which can contribute to declining unionization rates and potentially weaken the bargaining power of teachers’ unions. In 2011, Tennessee eliminated collective bargaining for public school teachers. In February 2025, Utah enacted legislation banning collective bargaining for all public sector workers. Given the constructive role unions have played during public emergencies, this recent wave of anti-union measures and increased legislative barriers may be steering the education system in a less favorable direction.

Due to the data limitations, I am unable to control for school and district characteristics beyond teachers’ school level. Because these factors may be correlated with both teachers’ labor market outcomes and unionization, I cannot claim that this study estimates the causal impact of teachers’ unions during the pandemic. Instead, this study identifies the impact of the COVID-19 pandemic on the difference in labor market performance between unionized and non-unionized teachers (i.e., returns on teacher unionization).

It is noteworthy that the analysis of the pandemic effect on the teachers’ labor market is based on teachers who are employed. Thus, the unemployed, the underemployed, and discouraged teachers are not included in the analysis. If the responses of these types of teachers are systematically different by union status, the results in this study may be biased in the presence of selectivity. Further analysis must be conducted to identify the mechanisms through which the pandemic influenced each of these teacher categories.

It is important to note that the CPS does not distinguish between union members covered by a collective bargaining agreement and those who are not, as it assumes that all union members are covered by collective bargaining agreements. This assumption is likely insufficient for public school teachers, because each state has its own labor laws regarding public sector employees. Moreover, excluding imputed earnings from the CPS can introduce several biases. For example, the imputation process may not fully reflect the characteristics of non-respondents, who are often more likely to be in lower income brackets; thus, their exclusion can lead to inflated average income estimates while understating poverty rates. Moreover, omitting imputed earnings may bias estimates of wage differentials, particularly when nonresponse rates differ across groups. Future research could utilize alternative micro-level datasets that offer more accurate information on the union status of public school teachers.